The Fed model, and the error inside it: known limitation 1

Using it to decide anything. The comparison has descriptive power, in that it explains how investors have actually priced equities against bonds, and that is a statement about behaviour rather than about value. A model that describes an error is not a reason to repeat it.

The Fed model, and the error inside it: known limitation 2

Quoting the spread as though it were a valuation measure with a record. Over the long US sample the earnings yield on its own carried the forecasting power, and subtracting the bond yield from it added essentially nothing, so the spread is the plain measure wearing an extra term that does no work.

Playbooks / Valuation

The Fed model, and the error inside it

Comparing an earnings yield with a government bond yield feels like comparing two prices for the same thing, and it is not, because one of them already contains inflation and the other does not.

The rule

The argument is everywhere and it sounds like arithmetic. Equities yield four percent on earnings, government bonds yield three, therefore equities are cheap. Or the reverse, and therefore expensive.

It is called the Fed model, on thin grounds, and the reason it persists is that it feels like a like for like comparison of two prices. It is not, and the reason it is not takes one sentence.

The error

The earnings yield is a real number. The bond yield is a nominal one.

Corporate earnings already contain inflation. When prices rise, revenues and earnings rise with them over time, so the earnings yield is a claim on a stream that grows. A government bond coupon is fixed in cash terms, so its yield already includes whatever inflation the market expects across its life.

Comparing the two directly treats a growing real claim and a fixed nominal one as though they were the same instrument at different prices. Asness gave this its clearest statement in 2003: the comparison is money illusion, and it has descriptive power precisely because investors make that error consistently.

That last part is worth separating out. The model does a reasonable job of describing how equities have actually been priced against bonds. It does that by capturing a mistake other people are making, and capturing a mistake is not the same as having a reason to make it yourself.

What the record shows

Over the long US sample, from 1926, the earnings yield on its own explained a meaningful share of subsequent ten year real equity returns, on the order of a third of the variation.

Subtracting the bond yield from it added essentially nothing.

So the spread has no forecasting record independent of the plain earnings yield. It is CAPE with a term bolted on that does no work, and the plain version was already a blunt instrument about decades rather than a signal about quarters.

The property version, which nobody calls the Fed model

Read this sentence, which appears in some form in most property research: cap rates look attractive relative to government bond yields, and the spread is wide by historical standards.

It is the identical comparison. A cap rate is a yield on net operating income, and net operating income grows roughly with inflation over long periods, imperfectly and with a lag. A ten year government yield is nominal and fixed. The spread between them moves when inflation expectations move, without anything happening to either the building or the credit of the government.

Which means the spread widens in exactly the environment where the nominal yield falls, and a narrative arrives to explain that property has become good value. Sometimes it has. The spread is not the evidence.

What to compare instead

If the point of the comparison is a hurdle, the hurdle should be real.

The real yield on an inflation protected government bond is the observable real return available on the safest asset there is, which inflation and real returns sets out. That number is directly comparable with a real earnings yield or a cap rate, because both sides are then measured in purchasing power.

The rest of the comparison is the risk premium, and naming it as a premium rather than burying it in a spread is what makes it arguable. Somebody who says property should yield three points above the real government yield has made a claim that can be discussed. Somebody who says the spread over nominal bonds is wide has not.

What the argument gets right

A framework that only attacks is not much use.

The cost of money does affect what an asset is worth, and the objection above is not a denial of that. A durable rise in real rates lowers what any income stream is worth, which is the mechanism gold and real rates turns on and the mechanism behind the part of the cap rate framework that hurts people.

The defect is specific. It is the use of a nominal yield on one side of a comparison whose other side is real. Fix that and the underlying idea survives; leave it and the conclusion moves whenever inflation expectations do, for reasons that have nothing to do with the asset being valued.

The arithmetic

The model as usually stated Earnings yield = E / P Compare with the ten year government bond yield. E/P above the bond yield equities called cheap E/P below the bond yield equities called expensive The objection E/P is a real quantity. Earnings already contain inflation, and grow with it. The bond yield is nominal. It contains expected inflation and does not grow. Comparing them treats the two as though they measured the same thing. What the evidence shows 1926 to 2001, forecasting the next ten years of real equity returns: E/P alone R2 about 35% E/P minus bond yield adds essentially nothing The property version Cap rate compared with the ten year yield, and called a spread. Same two quantities, same error. The fix Compare a real yield with a real yield, or state plainly that you are describing sentiment rather than measuring value.

Where it breaks

Using it to decide anything. The comparison has descriptive power, in that it explains how investors have actually priced equities against bonds, and that is a statement about behaviour rather than about value. A model that describes an error is not a reason to repeat it.
Quoting the spread as though it were a valuation measure with a record. Over the long US sample the earnings yield on its own carried the forecasting power, and subtracting the bond yield from it added essentially nothing, so the spread is the plain measure wearing an extra term that does no work.
Applying it to property as a cap rate spread over government bonds without noticing it is the same comparison. A cap rate is a yield on an income that grows roughly with inflation over long periods, and a government bond yield is a fixed nominal coupon, so a narrowing or widening spread between them is often just an inflation expectation moving.
Concluding that low rates justify any multiple. Lower nominal discount rates arrive with lower nominal cash flow growth, and the two effects offset each other to a degree the simple version of the argument ignores entirely.
Abandoning the comparison so completely that the real relationship between rates and asset prices is denied. Real yields do matter to what an asset is worth. The error is in the nominal version of the comparison, not in the idea that the cost of money is relevant.

When to use it

When somebody argues that equities or property are cheap because yields elsewhere are low, or expensive because yields have risen. It is a common argument, it is made in good faith, and the specific defect in it is worth being able to name.

Sources

Last reviewed 28 August 2026. Educational research, not personal advice. Disclosure standards.